Rapid Coupling Prediction for Compact GNSS Arrays via Graph Transformer

Abstract

The design of compact GNSS antenna arrays requires rapid evaluation of mutual coupling effects, which conventional full-wave simulations fail to deliver due to excessive computational costs. This letter proposes a graph transformer model that predicts coupling matrices in real time. By hybridizing the local perception of graph isomorphism network (GIN) and the global attention of transformer architectures, the method achieves physics-aware predictions with an average error of 2.81 dB, lower than conventional neural networks. And experimental results on a 16-element patch array show that the proposed model achieves inference in under 100 ms, compared to 60 min for CST simulations, enabling rapid exploration of next-generation GNSS receivers demanding miniaturization design.